Background: Mortality in HIV-infected people after initiation of antiretroviral treatment (ART) in resource-limited settings is an important measure of the effectiveness and comparative effectiveness of the global public health response. Substantial loss to follow-up precludes accurate accounting of deaths and limits our understanding of effectiveness. We aimed to provide a better understanding of mortality at scale and, by extension, the effectiveness and comparative effectiveness of public health ART treatment in east Africa. Methods: In 14 clinics in five settings in Kenya, Uganda, and Tanzania, we intensively traced a sample of patients randomly selected using a random number generator, who were infected with HIV and on ART and who were lost to follow-up (>90 days late for last scheduled visit). We incorporated the vital status outcomes for these patients into analyses of the entire clinic population through probability-weighted survival analyses. Findings: We followed 34 277 adults on ART from Mbarara and Kampala in Uganda, Eldoret, and Kisumu in Kenya, and Morogoro in Tanzania. The median age was 35 years (IQR 30-42), 11 628 (34%) were men, and median CD4 count count before therapy was 154 cells per μL (IQR 70-234). 5780 patients (17%) were lost to follow-up, 991 (17%) were selected for tracing between June 10, 2011, and Aug 27, 2012, and vital status was ascertained for 860 (87%). With incorporation of outcomes from the patients lost to follow-up, estimated 3 year mortality increased from 3·9% (95% CI 3·6-4·2) to 12·5% (11·8-13·3). The sample-corrected, unadjusted 3 year mortality across settings was lowest in Mbarara (7·2%) and highest in Morogoro (23·6%). After adjustment for age, sex, CD4 count before therapy, and WHO stage, the sample-corrected hazard ratio comparing the settings with highest and lowest mortalities was 2·2 (95% CI 1·5-3·4) and the risk difference for death at 3 years was 11% (95% CI 5·0-17·7). Interpretation: A sampling-based approach is widely feasible and important to an understanding of mortality after initiation of ART. After adjustment for measured biological drivers, mortality differs substantially across settings despite delivery of a similar clinical package of treatment. Implementation research to understand the systems, community, and patients' behaviours driving these differences is urgently needed. Funding: The US National Institutes of Health and President's Emergency Fund for AIDS Relief.